A key feature of decentralization in developing countries has been the creation of new local governments. The implications of this process for violent conflict are not well understood. On the one hand, bringing representative government closer to the electorate can reduce heterogeneity in preferences, thereby mitigating conflict. On the other hand, creating local government institutions also leads to a large increase in rents that may be contested violently. Group cleavages can determine which of these two effects prevails. Identifying these distinct channels empirically has proven difficult. This paper resolves these challenges by exploiting a natural experiment in the ethnically and religiously diverse context of post-authoritarian Indonesia where rapid decentralization was accompanied by dramatic growth in the number of new districts and a resulting decline in ethnolinguistic fractionalization. We use new microdata on conflict from 2000–2014 and leverage the plausibly exogenous timing of redistricting due to a government moratorium. Overall, redistricting has small and insignificant average effects on conflict. However, areas that experience greater ethnolinguistic and religious homogenization as a result of splitting experience a significant reduction in conflict. At the same time, we find a differential increase in violence in areas that receive a new government and are also ethnically polarized. These differential increases in violence are most pronounced around the time of the first election and for types of violence associated with contestation of public resources and institutions. These results suggest that allowing for redistricting along group lines can reduce conflict, but the benefits of reduced diversity may be undone if the newly governed population is highly polarized. In such cases, conflict may then simply shift from the original seats of government to newly created ones.